问题描述
我在下面的CTE查询中用“删除”替换“选择*”时遇到麻烦。我没有提供示例表，因为我认为/希望可以不给出答案。问题是如何删除这些返回的记录而不是选择它们-因为当我输入delete时出现错误：
“意外删除”
如何删除这些记录？请注意，当我在最底部使用“选择*”时，效果很好。
with block_1 as
(
select *
from table_1
where col_b is null
and col4 || col3 in (select col5 || col6
from table1
group by 1
having count(*) > 1)
),block_2 as
(
select *
from table1
where col_b is not null
and col4 || col3 in (select col5 || col6
from table_1
group by 1
having count(*) > 1)
)
select *
from table_1
where col4 || col3 in (select col5 || col6
from table1
group by 1
having count(*) > 1)
and misc_id in (select distinct misc_id from table1
where misc_id in (select misc_id from block_1)
and misc_id in (select misc_id from block_2))
and col_b is null;
解决方法
不可能同时使用CTE和DELETE操作。
解决方案将重写您的查询，例如使用USING子句避免CTE。
您可以在is there an alternative to query with DELETE snowflake SQL statement with CTE?和https://docs.snowflake.com/en/sql-reference/sql/delete.html这里找到更多信息
,因此，如果我们将SELECT CTE从IN形式转换为JOIN形式（我们可以这样做，因为IN中的SUB-SELECT提取不同的值），我们可以将所有代码转换为该形式。我暂时在main_s1中留了两个IN。
WITH block_1_s1 AS (
SELECT col5 || col6 AS col56
FROM table1
GROUP BY 1
HAVING COUNT(*) > 1
),block_1 AS (
SELECT t1.misc_id
FROM table_1 AS t1
JOIN block_1_s1 AS t2
ON t2.col56 = t1.col4 || t1.col3
WHERE t1.col_b IS NULL
),block_2_s1 AS (
SELECT col5 || col6 as col56
FROM table_1
GROUP BY 1
HAVING COUNT(*) > 1
),block_2 AS (
SELECT t1.misc_id
FROM table1 AS t1
JOIN block_2_s1 as t2
ON t2.col56 = t1.col4 || t1.col3
WHERE t1.col_b IS NOT NULL
),main_s1 AS (
SELECT DISTINCT misc_id
FROM table1
WHERE misc_id IN (select misc_id from block_1)
AND misc_id IN (select misc_id from block_2)
)
SELECT t1.*
FROM table_1 AS t1
JOIN block_1_s1 AS t2
ON t2.col56 = t1.col4 || t1.col3
JOIN main_s1 as t3 ON t3.misc_id = t1.misc_id
WHERE t1.col_b is null;
然后在main_s1 CTE中枢转两个IN，我们需要在misc_id上强制唯一性/区别性，以便我们可以从IN（转到JOIN）
WITH block_1_s1 AS (
SELECT col5 || col6 AS col56
FROM table1
GROUP BY 1
HAVING COUNT(*) > 1
),block_1 AS (
SELECT DISTINCT t1.misc_id
FROM table_1 AS t1
JOIN block_1_s1 AS t2
ON t2.col56 = t1.col4 || t1.col3
WHERE t1.col_b IS NULL
),block_2 AS (
SELECT DISTINCT t1.misc_id
FROM table1 AS t1
JOIN block_2_s1 as t2
ON t2.col56 = t1.col4 || t1.col3
WHERE t1.col_b IS NOT NULL
),main_s1 AS (
SELECT DISTINCT t1.misc_id
FROM table1 AS t1
JOIN block_1 AS b1 ON b1.misc_id = t1.misc_id
JOIN block_2 AS b2 ON b2.misc_id = t1.misc_id
)
SELECT t1.*
FROM table_1 AS t1
JOIN block_1_s1 AS t2
ON t2.col56 = t1.col4 || t1.col3
JOIN main_s1 as t3 ON t3.misc_id = t1.misc_id
WHERE t1.col_b is null;
因此现在对SQL进行了重复数据删除，并将所有数据都转换为JOIN（我们实际上可以IN形式完成此操作，就像JOINs一样。），我们可以将所有CTE替换为子选择：
SELECT t1.*
FROM table_1 AS t1
JOIN (
SELECT col5 || col6 AS col56
FROM table1
GROUP BY 1
HAVING COUNT(*) > 1
) AS t2
ON t2.col56 = t1.col4 || t1.col3
JOIN (
SELECT DISTINCT t1.misc_id
FROM table1 AS t1
JOIN (
SELECT DISTINCT t1.misc_id
FROM table_1 AS t1
JOIN (
SELECT col5 || col6 AS col56
FROM table1
GROUP BY 1
HAVING COUNT(*) > 1
) AS t2
ON t2.col56 = t1.col4 || t1.col3
WHERE t1.col_b IS NULL
) AS b1
ON b1.misc_id = t1.misc_id
JOIN (
SELECT DISTINCT t1.misc_id
FROM table1 AS t1
JOIN (
SELECT col5 || col6 as col56
FROM table_1
GROUP BY 1
HAVING COUNT(*) > 1
) AS t2
ON t2.col56 = t1.col4 || t1.col3
WHERE t1.col_b IS NOT NULL
) AS b2
ON b2.misc_id = t1.misc_id
) as t3
ON t3.misc_id = t1.misc_id
WHERE t1.col_b is null;
现在我们可以从SELECT切换到DELETE了。
我从IN表单选择JOIN的主要原因是Snowflake的排序/合并比基于Row的更多，而IN是JOIN的一种特殊形式，如果您有索引（Snowflake没有），可以快速。因此，鉴于我们确实在链接JOIN，我们在生产环境中发现了这些联接，您可以在这些联接中预先区分出前几个阶段执行IN，即使它们是相同的（当数据被强制用于JOIN时也是安全的），但是当优化编译器无法看到JOIN表单的速度超过了IN（在配置文件窗口中没有变成JOIN表单）。